CS341 Final Report: Towards Real-time Detection and Camera Triggering

In this project, we aim at deploying a real-time object detection system that operates at high FPS on resourceconstrained device such as Raspberry Pi and mobile phones. Although many systems have proved their success since the era of machine learning and neural network, most of the evaluations are done with high-end CPU or GPU. Nevertheless, real-world applications such as surveillance pose strict constraints on the resources of the device: lowpower, small form factor, relatively good accuracy and fast speed, making it hard to find a good trade-off when designing the system. We present an object detection pipeline which is capable of working smoothly under the situation of traffic surveillance on Raspberry Pi 3 with only 1GB RAM and 1.2GHz ARM CPU that costs merely $35.99.